#1 on Terminal Bench on 09/10

The agent that built itself.

A self-improving coding agent that handles the full dev lifecycle, from PM to PR.

Where OB-1 is going

A system that gets
sharper every time it ships.

01

Auto-generates evals

Creates evaluation suites from past PRs, then climbs them. Every commit raises the bar for the next one.

02

Builds its own skills

Generates hooks, rules, and reusable skills from your codebase and session history. Gets sharper with every interaction.

03

Background agents

Safe sandboxed agents that keep working while you context-switch. Spawn exploration, planning, and execution in parallel.

04

Session forking & sharing

Fork any conversation for parallel exploration. Share sessions with a link so teammates can continue with full context.

05

PM mode

Never runs out of ideas. OB-1 proactively suggests next steps, identifies friction points, and plans entire workstreams autonomously.

06

Thread-linked PRs

Every conversation thread produces a fully-contextualized pull request. Reviewers see the reasoning, not just the diff.

Integrations

Lives where you
already work.

Slack, Linear, GitHub, Graphite. OB-1 embeds directly into your workflow. No context-switching. Delegate tasks as they come to mind.

View all integrations

GitHub

PRs, issues, code review

"Fix the auth bug in issue #412" — OB-1 opens a PR in minutes.

Linear

Issue tracking and triage

Syncs issues, auto-triages, and updates status as work ships.

Graphite

Stacked PRs and merge queues

Splits large changes into reviewable stacked PRs automatically.

Slack

Delegate from any channel

Type /ob1 fix auth bug in any channel.

VS Code

IDE extension support

Cmd+Shift+O to hand off the current file to OB-1.

Terminal

Native CLI agent

ob1 "refactor the auth module" — ships from your terminal.

FAQS

Frequently asked questions

OB-1 is a terminal-based coding agent — like Claude Code, but with multi-model access and self-optimizing workflows. You bring your own API keys (or use ours), pick the best model for each task, and let OB-1 handle the rest.

OB-1 learns from your sessions. It builds skills from your codebase patterns, auto-discovers workflow optimizations via /inbox, and spawns parallel subagents that work independently. Session forking lets you explore alternatives without losing your main thread.

300+ models via OpenRouter — Claude (including Opus 4.6), GPT, Gemini, Grok, Qwen, MiniMax, Llama, and more. Set your model to auto and OB-1 picks the best one for each prompt. Fast inference through Cerebras and SambaNova for open-source models.

GitHub, Linear, Slack, Graphite, VS Code, and any MCP-compatible tool. OB-1 also has a built-in browser agent for web interaction, persistent Chrome profiles for authenticated workflows, and verifiers that auto-run your build/test/lint commands after every change.

Up and running in minutes.

Install, point at your repo, and let OB-1 handle the rest — zero config required.